⚑No verdict, no hierarchy
Nine outcome cards render at equal visual weight. Nothing tells the operator the quarter is fine except one dragging pillar — they must derive the headline themselves, every time.
FIX → a plain-language headline verdict + three stat callouts up top
⚑Insight without action
Drivers and forces are shown, but there is no ranked list of what to do first. “No-shows are up” sits next to “taste coverage rising” with no priority between them.
FIX → a ranked act-now queue with DO FIRST + impact chips
⚑Filter-first, not answer-first
Seven dropdown filters greet the operator before a single answer. The tool asks the user to do the query instead of leading with the finding.
FIX → scorecard leads; filtering demoted, drill on click
⚑Evidence is uncited
Numbers appear with no provenance. An operator can’t tell which pillar or endpoint produced “+19% demand”, so they can’t trust or trace it.
FIX → every insight card cites its real Atlas endpoint
⚑Decisions have nowhere to land
When two guest profiles are the same person, or a driver demands a call, the page can surface it but offers no place to resolve it — the operator leaves to act elsewhere.
FIX → inline merge-decision panel with an idempotent action